Radiomic signatures as predictors of pathological response to neoadjuvant chemoimmunotherapy in surgically resected NSCLC.

Abstract

8080 Background: Historically, pathological complete response (pCR), a potential early predictor of survival, was achieved by a small fraction of patients with non-small cell lung cancer (NSCLC) receiving neoadjuvant chemotherapy. Now with chemoimmunotherapy (chemo-IO) becoming the cornerstone of perioperative treatment, the rate of pCR has significantly increased to over 15%. Existing factors like PD-L1 expression and circulating tumor DNA clearance have shown limited efficacy in reliably predicting response to neoadjuvant chemo-IO, thus underscoring the need for novel biomarkers. In this study, we aim to investigate the potential of radiomic texture features derived from pre-treatment CT scans to predict pCR in patients with NSCLC undergoing neoadjuvant chemo-IO prior to surgery. Methods: The study included 101 patients with surgically resected NSCLC treated at Cleveland Clinic. All patients received neoadjuvant platinum-doublet chemotherapy combined with an anti-PD-1 inhibitor prior to surgery. Tumor stage, histology, PD-L1 expression levels, and treatment details (e.g., chemotherapy regimen, immunotherapy agent, number of treatment cycles) were collected for analysis. Pathological responses were assessed based on the percentage of residual viable tumor in the surgical specimen, with pathological complete response (pCR) defined as 0% viable tumor. Radiomic features were extracted from both intratumoral and peritumoral regions on pre-treatment CT images. Patients were randomly divided into training and validation cohorts, ensuring an equal distribution of pCR and non-pCR cases in the training set. The training cohort (St) comprised 50 patients, while the validation cohort (Sv) included 51 patients. A linear discriminant classifier (LDA) was trained using St and subsequently evaluated on Sv. The predictive performance was assessed using the area under the curve (AUC). Results: 37 of 101 patients (37%) achieved a pCR. Utilizing a combination of 5 peritumoral and intratumoral radiomic features extracted from pretreatment CT scans, the AUC for predicting pCR was 0.82 (95% CI: 0.79 − 0.86) in St and 0.78 (95% CI: 0.76 − 0.81) in Sv. In contrast, the predictive capability of PD-L1 expression alone yielded an AUC of 0.57 for pCR prediction. Moreover, no significant difference was observed in pCR rates between patients with low and high PD-L1 expression levels (P = 0.1). The integration of radiomic features with clinicopathologic factors, including age, race, tumor stage, and PD-L1 expression, resulted in a modest improvement in predictive performance (AUC = 0.8) but was not statistically significant (P > 0.5). Conclusions: This analysis suggests that radiomic features extracted from both intra- and peri-tumoral regions on pre-treatment CT images may be indicative of the probability of achieving a pCR in patients with NSCLC receiving neoadjuvant chemo-IO.